We present the first version of a semantic reasoning benchmark for Danish compiled semi-automatically from a number of human-curated lexical-semantic resources, which function as our gold standard. Taken together, the datasets constitute a benchmark for assessing selected language understanding capacities of large language models (LLMs) for Danish. This first version comprises 25 datasets across 6 different tasks and include 3,800 test instances. Although still somewhat limited in size, we go beyond comparative evaluation datasets for Danish by including both negative and contrastive examples as well as low-frequent vocabulary; aspects which tend to challenge current LLMs when based substantially on language transfer. The datasets focus on features such as semantic inference and entailment, similarity, relatedness, and ability to disambiguate words in context. We use ChatGPT to assess to which degree our datasets challenge the ceiling performance of state-of-the-art LLMs, average performance being relatively high with an average accuracy of 0.6 on ChatGPT 3.5 turbo and 0.8 on ChatGPT 4.0.
In this paper, we present the newly compiled DA-ELEXIS Corpus, which is one of the largest sense-annotated corpora available for Danish, and the first one to be annotated with the Danish wordnet, DanNet. The corpus is part of a European initiative, the ELEXIS project, and has corresponding parallel annotations in nine other European languages. As such it functions as a cross-lingual evaluative benchmark for a series of low and medium resourced European language. We focus here on the Danish annotation process, i.e. on the annotation scheme including annotation guidelines and a primary sense inventory constituted by DanNet as well as the fall-back sense inventory namely The Danish Dictionary (DDO). We analyse and discuss issues such as out of vocabulary (OOV) problems, problems with sense granularity and missing senses (in particular for verbs), and how to semantically tag multiword expressions (MWE), which prove to occur very frequently in the Danish corpus. Finally, we calculate the inter-annotator agreement (IAA) and show how IAA has improved during the annotation process. The openly available corpus contains 32,524 tokens of which sense annotations are given for all content words, amounting to 7,322 nouns, 3,099 verbs, 2,626 adjectives, and 1,677 adverbs.
Systematic polysemy is a well-known linguistic phenomenon where a group of lemmas follow the same polysemy pattern. However, when compiling a lexical resource like a wordnet, a problem arises regarding when to underspecify the two (or more) meanings by one (complex) sense and when to systematically split into separate senses. In this work, we present an extensive analysis of the systematic polysemy patterns in Danish, and in our preliminary study, we examine a subset of these with experiments on human intuition and contextual embeddings. The aim of this preparatory work is to enable future guidelines for each polysemy type. In the future, we hope to expand this approach and thereby hopefully obtain a sense inventory which is distributionally verified and thereby more suitable for NLP.
In this paper we report on a new Danish lexical initiative, the Central Word Register for Danish, (COR), which aims at providing an open-source, well curated and large-coverage lexicon for AI purposes. The semantic part of the lexicon (COR-S) relies to a large extent on the lexical-semantic information provided in the Danish wordnet, DanNet. However, we have taken the opportunity to evaluate and curate the wordnet information while compiling the new resource. Some information types have been simplified and more systematically curated. This is the case for the hyponymy relations, the ontological typing, and the sense inventory, i.e. the treatment of polysemy, including systematic polysemy.
We present The Central Word Register for Danish (COR), which is an open source lexicon project for general AI purposes funded and initiated by the Danish Agency for Digitisation as part of an AI initiative embarked by the Danish Government in 2020. We focus here on the lexical semantic part of the project (COR-S) and describe how we – based on the existing fine-grained sense inventory from Den Danske Ordbog (DDO) – compile a more AI suitable sense granularity level of the vocabulary. A three-step methodology is applied: We establish a set of linguistic principles for defining core senses in COR-S and from there, we generate a hand-crafted gold standard of 6,000 lemmas depicting how to come from the fine-grained DDO sense to the COR inventory. Finally, we experiment with a number of language models in order to automatize the sense reduction of the rest of the lexicon. The models comprise a ruled-based model that applies our linguistic principles in terms of features, a word2vec model using cosine similarity to measure the sense proximity, and finally a deep neural BERT model fine-tuned on our annotations. The rule-based approach shows best results, in particular on adjectives, however, when focusing on the average polysemous vocabulary, the BERT model shows promising results too.
This paper describes how a newly published Danish sentiment lexicon with a high lexical coverage was compiled by use of lexicographic methods and based on the links between groups of words listed in semantic order in a thesaurus and the corresponding word sense descriptions in a comprehensive monolingual dictionary. The overall idea was to identify negative and positive sections in a thesaurus, extract the words from these sections and combine them with the dictionary information via the links. The annotation task of the dataset included several steps, and was based on the comparison of synonyms and near synonyms within a semantic field. In the cases where one of the words were included in the smaller Danish sentiment lexicon AFINN, its value there was used as inspiration and expanded to the synonyms when appropriate. In order to obtain a more practical lexicon with overall polarity values at lemma level, all the senses of the lemma were afterwards compared, taking into consideration dictionary information such as usage, style and frequency. The final lexicon contains 13,859 Danish polarity lemmas and includes morphological information. It is freely available at https://github.com/dsldk/danish-sentiment-lexicon (licence CC-BY-SA 4.0 International).
The paper describes work in progress in the DanNet2 project financed by the Carlsberg Foundation. The project aim is to extend the original Danish wordnet, DanNet, in several ways. Main focus is on extension of the coverage and description of the adjectives, a part of speech that was rather sparsely described in the original wordnet. We describe the methodology and initial work of semi-automatically transferring adjectives from the Danish Thesaurus to the wordnet with the aim of easily enlarging the coverage from 3,000 to approx. 13,000 adjectival synsets. Transfer is performed by manually encoding all missing adjectival subsection headwords from the thesaurus and thereafter employing a semi-automatic procedure where adjectives from the same subsection are transferred to the wordnet as either 1) near synonyms to the section’s headword, 2) hyponyms to the section’s headword, or 3) as members of the same synset as the headword. We also discuss how to deal with the problem of multiple representations of the same sense in the thesaurus, and present other types of information from the thesaurus that we plan to integrate, such as thematic and sentiment information.
Aligning senses across resources and languages is a challenging task with beneficial applications in the field of natural language processing and electronic lexicography. In this paper, we describe our efforts in manually aligning monolingual dictionaries. The alignment is carried out at sense-level for various resources in 15 languages. Moreover, senses are annotated with possible semantic relationships such as broadness, narrowness, relatedness, and equivalence. In comparison to previous datasets for this task, this dataset covers a wide range of languages and resources and focuses on the more challenging task of linking general-purpose language. We believe that our data will pave the way for further advances in alignment and evaluation of word senses by creating new solutions, particularly those notoriously requiring data such as neural networks. Our resources are publicly available at https://github.com/elexis-eu/MWSA.
Although Denmark is one of the most digitized countries in Europe, no coordinated efforts have been made in recent years to support the Danish language with regard to language technology and artificial intelligence. In March 2019, however, the Danish government adopted a new, ambitious strategy for LT and artificial intelligence. In this paper, we describe the process behind the development of the language-related parts of the strategy: A Danish Language Technology Committee was constituted and a comprehensive series of workshops were organized in which users, suppliers, developers, and researchers gave their valuable input based on their experiences. We describe how, based on this experience, the focus areas and recommendations for the LT strategy were established, and which steps are currently taken in order to put the strategy into practice.
In this paper we describe the merge of the Danish wordnet, DanNet, with Princeton Wordnet applying a two-step approach. We first link from the English Princeton core to Danish (5,000 base concepts) and then proceed to linking the rest of the Danish vocabulary to English, thus going from Danish to English. Since the Danish wordnet is built bottom-up from Danish lexica and corpora, all taxonomies are monolingually based and thus not necessarily directly compatible with the coverage and structure of the Princeton WordNet. This fact proves to pose some challenges to the linking procedure since a considerable number of the links cannot be realised via the preferred cross-language synonym link which implies a more or less precise correlation between the two concepts. Instead, a subpart of the links are realised through near synonym or hyponymy links to compensate for the fact that no precise translation can be found in the target resource. The tool WordnetLoom is currently used for manual linking but procedures for a more automatic procedure in future is discussed. We conclude that the two resources actually differ from each other quite more than expected, both vocabulary and structure-wise.
Our aim is to develop principled methods for sense clustering which can make existing lexical resources practically useful in NLP – not too fine-grained to be operational and yet finegrained enough to be worth the trouble. Where traditional dictionaries have a highly structured sense inventory typically describing the vocabulary by means of mainand subsenses, wordnets are generally fine-grained and unstructured. We present a series of clustering and annotation experiments with 10 of the most polysemous nouns in Danish. We combine the structured information of a traditional Danish dictionary with the ontological types found in the Danish wordnet, DanNet. This constellation enables us to automatically cluster senses in a principled way and improve inter-annotator agreement and wsd performance.
We launch the SemDaX corpus which is a recently completed Danish human-annotated corpus available through a CLARIN academic license. The corpus includes approx. 90,000 words, comprises six textual domains, and is annotated with sense inventories of different granularity. The aim of the developed corpus is twofold: i) to assess the reliability of the different sense annotation schemes for Danish measured by qualitative analyses and annotation agreement scores, and ii) to serve as training and test data for machine learning algorithms with the practical purpose of developing sense taggers for Danish. To these aims, we take a new approach to human-annotated corpus resources by double annotating a much larger part of the corpus than what is normally seen: for the all-words task we double annotated 60% of the material and for the lexical sample task 100%. We include in the corpus not only the adjucated files, but also the diverging annotations. In other words, we consider not all disagreement to be noise, but rather to contain valuable linguistic information that can help us improve our annotation schemes and our learning algorithms.
In this article we present an expansion of the supersense inventory. All new super-senses are extensions of members of the current inventory, which we postulate by identifying semantically coherent groups of synsets. We cover the expansion of the already-established supernsense inventory for nouns and verbs, the addition of coarse supersenses for adjectives in absence of a canonical supersense inventory, and super-senses for verbal satellites. We evaluate the viability of the new senses examining the annotation agreement, frequency and co-ocurrence patterns.
This paper discusses how information on properties in a currently developed Danish thesaurus can be transferred to the Danish wordnet, DanNet, and in this way enrich the wordnet with the highly relevant links between properties and their external arguments (i.e. tasty ― food). In spite of the fact that the thesaurus is still under development (two thirds still to be compiled) we perform an automatic transfer of relations from the thesaurus to the wordnet which shows promising results. In all, 2,362 property relations are automatically transferred to DanNet and 2% of the transferred material is manually validated. The pilot validation indicates that approx. 90 % of the transferred relations are correctly assigned whereas around 10% are either erroneous or just not very informative, a fact which, however, can partly be explained by the incompleteness of the material at its current stage. As a further consequence, the experiment has led to a richer specification of the editor guidelines to be used in the last compilation phase of the thesaurus.
In this paper we investigate the problem of merging specialist taxonomies with the more intuitive folk taxonomies in lexical-semantic resources like wordnets; and we focus in particular on plants, animals and foods. We show that a traditional dictionary like Den Danske Ordbog (DDO) survives well with several inconsistencies between different taxonomies of the vocabulary and that a restructuring is therefore necessary in order to compile a consistent wordnet resource on its basis. To this end, we apply Cruses definitions for hyponymies, namely those of natural kinds (such as plants and animals) on the one hand and functional kinds (such as foods) on the other. We pursue this distinction in the development of the Danish wordnet, DanNet, which has recently been built on the basis of DDO and is made open source for all potential users at www.wordnet.dk. Not surprisingly, we conclude that cultural background influences the structure of folk taxonomies quite radically, and that wordnet builders must therefore consider these carefully in order to capture their central characteristics in a systematic way.
Topic: lexical ressources, international project Abstract: The LEXADV-project is a Scandinavian research project (2004-2006, financed by Nordplus Sprog) with the aim of extending three Scandinavian semantic lexicons building on the SIMPLE lexicon model (Lenci et al., 2000) with the word class of adverbs. In the lexicons of approx. 400 Danish, Norwegian and Swedish adverbs the different senses are described with a semantic type and a set of semantic features. A classification covering the many meanings that adverbs can have has been established and integrated in the original SIMPLE ontology. Similarly new features have been added to the model in order to describe the adverb senses. The working method of the project builds on the fact that the vocabularies of Danish, Norwegian and Swedish are closely related. An encoding tool has been developed with the special purpose of permitting easy transfer of semantic types and features between entries in the three languages. The Danish adverb senses have been described first, based on the definition in a modern, comprehensive Danish dictionary. Afterwards the lemmas have been translated and the semantic data have been copied into the Swedish as well as into the Norwegian equivalent entry. Finally these copies have been evaluated and when necessary adjusted by native speakers.
A word class often neglected in the field of NLP resources, namely adverbs, has lately been described in a computational lexicon produced at CST as one of the results of a Ph.D.-project. The adverb lexicon, which is integrated in the Danish STO lexicon, gives detailed syntactic information on the type of modification and position, as well as on other syntactic properties of approx 800 Danish adverbs. One of the aims of the lexicon has been to establish a clear distinction between syntactic and semantic information - where other lexicons often generalize over the syntactic behavior of semantic classes of adverbs, every adverb is described with respect to its proper syntactic behavior in a text corpus, revealing very individual syntactic properties. Syntactic information on adverbs is needed in NLP systems generating text to ensure correct placing in the phrase they modify. Also in systems analyzing text, this information is needed in order to attach the adverbs to the right node in the syntactic parse trees. Within the field of linguistic research, several results can be deduced from the lexicon, e.g. knowledge of syntactic classes of Danish adverbs.